Task Observer skill
task-observer - One Skill to Rule Them All
A meta-skill that watches the agent's work sessions and proposes improvements to existing skills and candidates for new ones
- 3,902 ★ on the repository
- Last commit
- CC-BY-4.0
Extended access
Why this level
- Reads work session context and writes observation logs to the filesystem
- Uses Python helper scripts to migrate logs and validate the skill bundle
Install
- Works in
Manual install
Download the .skill bundle from the latest release or clone the repo and place SKILL.md, references/ and scripts/ into one skills folder, for example .claude/skills/task-observer/. Then add the activation instruction from references/environments.md to CLAUDE.md and verify activation in a new session.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Reads work session context and writes observation logs to the filesystem
- Uses Python helper scripts to migrate logs and validate the skill bundle
This is third-party code. Review the repository files before installing.
What it does
The skill runs in the background during a normal session and watches three things: corrections and adjustments the user makes, work done manually that no skill covers yet, and its own blind spots. From these it keeps a structured observation log: what it noticed, which skills it affects and what specific improvements it suggests. Recurring patterns go into a separate log of cross-cutting principles that new skills are later checked against. The skill never edits anything itself; it prepares recommendations for review, and the decision stays with the user. In environments with filesystem access the log persists between sessions; without it the skill produces a handoff document at the end of the session.
Who it is for. For anyone who uses skills actively and wants their skill library to improve as they work rather than stay frozen.
Good fit when
- You want to improve a set of skills systematically from real work
- You want to spot repeating tasks and turn them into new skills
- You want an observation log and regular reviews for a large skill library
Not a fit when
- You have only a couple of skills and the agent's built-in memory plus direct editing is enough
- You are not ready to set up persistent activation and to review the observation log
Example request
Log any observations from this session and suggest which skills to improveLimitations
It pays off more as the skill library and workload grow; for a small setup the log and review overhead may not be worth it. Description matching alone is unreliable: you must add an activation instruction to CLAUDE.md or a session-start hook, otherwise the log never appears. A persistent log needs filesystem access and one stable path, otherwise observations are lost. The skill assumes Claude's architecture, and other platforms need to adapt some of its concepts.
How to disable. Remove the task-observer skill folder from the skills directory and delete the activation instruction from CLAUDE.md or the session-start hook. The observation logs can be removed separately.
FAQ
Does the skill edit my skills itself?
No. It keeps an observation log and prepares specific recommendations; whether to apply them is your call at review time.
Why is no log appearing?
The skill most likely is not activating. Description matching is not enough: add an activation instruction to CLAUDE.md or a session-start hook and verify in a new session.
Discussion
No questions or reviews yet.
Related
SkillsSelf-learning
self-learning-skills
A meta-skill that saves a hard-won working procedure as a new skill or rule, so the next session starts already knowing it
A skill for creating skills that proves them by behavior: before-and-after runs, regression evals and ecosystem audit
A doctor skill for agent skills that picks an architecture, audits against 21 red lines and returns a report with line references
The xskill CLI gathers sanitized session trajectories from Claude Code, Codex and OpenCode and grows a shared skill library.
A meta-skill that turns a workflow into an agent skill: prior-art search, trigger evaluation and publication to GitHub
A workshop skill: the agent vets a finished skill, finds peers, scores it and edits step by step, testing every change.
README in short
The README introduces a meta-skill that watches work sessions, catches user corrections, uncovered tasks and its own gaps, and turns them into skill improvements and candidates for new skills. It changes nothing itself; it prepares recommendations for review and keeps an observation log plus a separate log of cross-cutting principles. Installation is via the .skill release bundle or a folder with SKILL.md, references/ and scripts/, and activation must be enabled separately through CLAUDE.md or a hook and verified in a new session. The skill targets Claude environments and falls back to a handoff document where there is no filesystem. CC BY 4.0 licensed.
README badge
Original SKILL.md
--- name: task-observer description: > Monitors task execution for skill improvement opportunities. Use during ANY multi-step task, agentic workflow, or work session where the agent uses tools and produces deliverables. Captures patterns, user corrections, workflow insights, and methodology worth preserving as reusable skills. Also triggers in post-task feedback discussions and when the user mentions skill observations, improvements, the observation log, skill taxonomy, or asks the agent to watch for skill opportunities. Also known as "One Skill to Rule Them All" — trigger on this phrase too. IMPORTANT: invoke this skill before the FIRST tool call of any session and before writing or proposing a plan. --- # Task Observer — Continuous Skill Discovery & Improvement Skills improve best from friction noticed during real work, not from sitting down to "improve a skill." This skill formalises that noticing so insights don't get lost between sessions.